For fifty years Eurovision drew its running order by lot. In those finals, a song drawn into the last third of the night finished about one and a half places higher than one drawn into the first.
A lottery is an experiment nobody had to design: no producer chose who went late, so if late songs did better, it was the slot and not the song. Every figure below is recalculated in your browser from 2,148 entries in 107 shows, 1957 to 2025, extracted from each contest's own Wikipedia article. And the famous curse of slot two? That is what chance leaves somewhere, and slot sixteen has never won either.
Slot by slot
Every final whose order was drawn by lot, 1957 to 2006. Each bar is where songs from that tenth of the running order finished, as a share of the way down the scoreboard: lower is better. The grey band is the middle 95% of what the same finals produce when the slots are shuffled at random, two thousand times.
Hover a bar.
computing…
Two measures, never averaged
A finishing place and a qualification are different events: different nights, different songs, different voters. If running order matters, both should show it. They are kept apart on purpose, so that if one moved and the other did not, you would see it.
Measure A · finals drawn by lot
Where they finished
Mean place, by third of the running order, 1957–2006. Taller is a better place.
computing…
Measure B · semi-finals drawn by lot
Who got through
Share that qualified for the final, by third, 2004–2012, choosers removed.
computing…
The control
If the hat really was a hat, a country's slot should know nothing about how good that country was before the draw. So the same test is run on something the running order cannot have caused: each country's average finish over its previous five finals. If that leans the way the result leans, the draw was not clean, and the headline above would not stand.
computing…
The curse of slot two
No song performed second has won Eurovision. It is repeated every May. Here are the winners of all 68 finals by slot, under every running-order regime, with ties counted for each winner.
Hover a slot.
Rerun history
If the winner were drawn out of the hat too
Each run replays all the finals with the same field sizes, giving every song the same chance of winning, then checks the slots that have been used in at least 40 finals.
The hat kept changing
The result depends entirely on knowing which years were a lottery, and the rule was not constant. These are the regimes, as each contest's own article describes its draw. Getting this wrong would quietly mix producer decisions into an experiment.
1956
Seven countries, two songs each. Only the winner was ever announced, so there are no places to measure. Left out.
1957–2006
The running order is drawn by lot, typically months ahead (in 1961 a six-year-old and a four-year-old helped). Countries that withdraw after the draw leave a gap the remaining entries close up around.
1996
An audio-only qualifying round: juries heard 29 songs on tape, in a drawn order, and 22 went through. No stage at all.
2007
Five semi-final countries and three finalists win the right to choose their slot. All eight pick the second half of the show. The people with the most at stake believed the folk wisdom.
2008
Again some countries are drawn to pick their own position: three in each semi-final, and Serbia, the host, in the final.
2009–2012
The articles describe a running-order draw and say nothing about choosers either way. Kept with 2007–08 as "lot with exceptions", out of the headline.
2013–
Countries draw only which half they perform in; the show's producers set the exact order. No longer an experiment.
The same test, regime by regime
Methods
Sources
English Wikipedia, "Eurovision Song Contest 1957" to "Eurovision Song Contest 2025": 69 articles read through the MediaWiki parse API on 2026-10-07. Every result table (running order, country, points, place) and each article's account of how the running order was set. Text and tables are licensed CC BY-SA 4.0; the exact revision of each article is recorded in the repository, so the numbers here can be checked against what was read. The derived data on this page is shared under the same licence.
No other data. The original plan was to use a widely shared Eurovision dataset on GitHub; it carries no licence and was scraped from a fan site, so it was set aside and the tables were rebuilt from the articles instead.
What was measured, and how
Slot is the entry's rank in the show's running order, so a country that withdrew after the draw does not leave a hole. The Netherlands in 2024 was disqualified before the final and holds no slot. Position within the night is slot minus one over field size minus one: 0 for the opener, 1 for the closer.
Measure A is the finishing place as a share of the way down the scoreboard, (place − 1) / (field − 1), so a field of 16 and a field of 26 sit on the same scale. Ties keep the place the article gives them (1969 had four winners).
Measure B is whether a semi-final entry reached that year's final. It is derived from the final's own table, not from row colouring, and every semi-final checks out at exactly ten qualifiers.
Significance is a permutation test: slots are shuffled within each show, 2,000 times, and the observed correlation is compared with that spread. Shuffling within a show keeps every year's field size and scoring system intact.
The control is each country's mean Measure A over its previous five finals (at least two needed), correlated with its slot in the same way. Its verdict is computed live: if it ever falls outside its own chance band, the headline figure is replaced by a warning.
The places figure in the headline is the gap in mean (place − 1) / (field − 1) between the first and last thirds, multiplied by the median field size minus one.
Parse checks: in every one of the 107 tables, more points never comes with a worse place, and every semi-final yields ten qualifiers. Both checks are in src/shape.py and stop the build if they fail.
What this does not say
Not why. Juries tiring, televoters remembering the last few songs, the long tail of early slots being used for warm-up acts: the lottery says the slot matters, not which of these is the reason.
Nothing causal about producer years. Since 2013 producers place the songs, and they have said they do it to pace the show. The 2013–25 numbers are shown for comparison only; a song placed late by a producer may simply be a song the producer rated.
Not about any one song. The effect is about one and a half places on average, against a spread of the whole scoreboard. Plenty of openers have won: three in the lottery years.
Not a ruling on slot two. Slot two has no winners, and that is unremarkable in aggregate. It does not show slot two is fine either; with this few finals, no single slot's record can show much.
Not the whole scoring history. Juries, televotes and points scales changed many times. Working in shares of the field keeps years comparable, but it does not make a 1965 jury and a 2005 televote the same instrument.
Wikipedia is the only source. The tables are well kept and pass both parse checks, but an error in an article is an error here. Each figure can be traced to a dated revision.
What did not work
2009–2012 could not be classified. 2007 and 2008 name the countries that chose their slots, so they can be removed. For 2009–12 the articles describe a draw and are silent on whether anyone chose. Rather than guess, those years sit in their own regime and stay out of the headline. Including them makes the effect larger, not smaller, which is a reason for caution and not for comfort.
The no-stage control is too small to use. The 1996 qualifying round was judged from tapes in a drawn order, which should separate "being heard late" from "being seen late". It has 29 songs. Its result is shown here for completeness, but a single round cannot tell those apart, and it is not used for anything.
The 2007 choosers are evidence, but not data. Eight countries choosing the second half shows what delegations believed. It cannot be tested, because the countries chose; those eight entries are excluded from every test unless you switch them back in above.